REANNZ develops and supports a range of products and services to support the specialist needs of our members in the R&E and innovation community.
We operate NZ's national research and education network and seamlessly connect with 120+ networks globally to enable researchers to collaborate.
REANNZ is proud to support the specialist needs of our members from NZ's research, education and innovation community.
Find out more about who REANNZ is, what we do and the people that operate New Zealand’s national research and education network.
Researchers are developing an artificial intelligence-based platform for the early prediction of cerebral palsy and other neurological disorders.
Movement analysis of an infant at an early age can reveal beneficial information about the child’s brain and nervous system and, more importantly, whether the baby is at risk of developing neurological disorders.
Early detection of such neurodevelopmental issues, for example, those that might lead to cerebral palsy, provides us the opportunity to intervene and reduce the severity of the symptoms.
The most predictive assessment of early neurodevelopment is the General Movements Assessment (GMA), in which clinicians manually classify videos of supine, spontaneously moving infants (9–16 weeks corrected age), but this assessment is limited in scalability by the availability of trained assessors.
At the University of Auckland, researchers Angus McMorland, Hamid Abbasi, and Manpreet Kaur are developing an artificial intelligence-based platform for the early prediction of cerebral palsy and other neurological disorders through the automatic analysis of video recordings of babies during their early months of life.
This fully automated, machine-learning-based implementation of the GMA could be scaled nationwide and may provide increased sensitivity by virtue of being a quantitative analysis that can be trained against long-term neurological outcomes.
Ultimately, the project aims to permit widespread access to the GMA, which will provide early identification of at-risk infants and thus early access to treatment that will mitigate the severity of life-long disability.
Chris Scott, Maxime Rio, and Dinindu Senanayake, Research Software Engineers who worked with New Zealand eScience Infrastructure at the time, collaborated with the scientists to deploy the DeepLabCut and lightning-pose software.
The software was containerised using Apptainer, where each container includes all required dependencies and is completely self-contained, resulting in a more reproducible build that can easily be moved to other platforms and is less likely to have issues related to dependency version mismatches.
This is particularly important for machine learning software such as TensorFlow and pytorch, which can have very specific version requirements on GPU-based dependencies such as CUDA and cuDNN.
Training these models can take a long time even with access to fast GPUs, with the training sometimes running for weeks. On a shared facility like the REANNZ HPC Platform, there are limits to how long you can run jobs for.
We investigated checkpointing and resuming the training process, allowing the training to be stopped before completion and then resumed from where it left off, rather than having to complete the entire training in one go. This also makes the training more robust as node failures can happen and with checkpoints you don’t lose everything if the node fails while your job is running on it.
A prototype backend system was developed that provides an API for uploading participant data, including videos of the infants moving. A mobile application, developed separately for data capture, integrates with this API to provide a convenient interface for clinicians and parents. The backend system was developed on our Research Developer Cloud, a flexible cloud platform that sits next to the national HPC Platform.
The backend software was containerised and deployed to again support reproducible deployments and portability, as well as account for the need to encrypt data and to manage the secrets used in the encryption.
The current backend system is an early prototype and further work would be required to integrate this with the machine learning models being developed. An example of a future integration includes automatically sending newly uploaded videos to be processed by the machine learning model and return the results.
The expertise and eagerness to find solutions for us by the team was absolutely top-rate. They have created, to a very professional standard, critical infrastructure for our project that securely receives and hosts sensitive patient information right where we need it near the high-performance capabilities of the HPC platform. The team was incredibly friendly, knowledgeable, and accommodating, and were a pleasure to work with. This system they have created will be instrumental in our endeavour to improve the lives of some of our most vulnerable infants.
Angus McMorland, Exercise, Sport & Rehab Science, University of Auckland
Find anything about our products, services, and more. Enter a query in the search input above.
